Code for our paper: "Knowledge Distillation During Mid-Training Favors Reasoning over Factual Recall" - View it on GitHub